"""
Rules-based intelligence extraction from call transcripts.

Scans a transcript for known incident keywords, categorises the call, and
extracts rough location hints (street/intersection mentions).

No external ML dependencies — fast and always available even when STT is
disabled. Designed to run as part of the post-upload background pipeline.
"""
import re
from typing import Optional
from app.internal.logger import logger
from app.internal import firestore as fstore


# ---------------------------------------------------------------------------
# Keyword taxonomy
# ---------------------------------------------------------------------------

INCIDENT_KEYWORDS: dict[str, list[str]] = {
    "fire": [
        "fire", "smoke", "flames", "burning", "structure fire", "brush fire",
        "wildfire", "arson", "working fire", "fully involved",
    ],
    "ems": [
        "cardiac", "unconscious", "breathing", "overdose", "trauma",
        "injury", "ambulance", "ems", "medic", "chest pain", "stroke",
        "unresponsive", "fall", "laceration",
    ],
    "police": [
        "pursuit", "chase", "shots fired", "weapon", "suspect", "robbery",
        "assault", "burglary", "stolen", "fleeing", "armed", "shooting",
        "stabbing", "domestic",
    ],
    "accident": [
        "accident", "collision", "crash", "mvr", "vehicle", "rollover",
        "hit and run", "ped", "pedestrian", "pi", "property damage",
    ],
}

# Street suffix patterns for location extraction
_STREET_RE = re.compile(
    r'\b(?:\d+\s+)?[A-Z][a-zA-Z]+(?: [A-Z][a-zA-Z]+)*'
    r'\s+(?:Street|St|Avenue|Ave|Boulevard|Blvd|Drive|Dr|Road|Rd|Lane|Ln'
    r'|Court|Ct|Place|Pl|Way|Circle|Cir|Highway|Hwy|Route|Rt)\b',
    re.IGNORECASE,
)


# ---------------------------------------------------------------------------
# Public API
# ---------------------------------------------------------------------------

async def extract_tags(
    call_id: str,
    transcript: str,
) -> tuple[list[str], Optional[str]]:
    """
    Extract incident tags from a transcript.

    Returns:
        (tags, primary_type) — e.g. (["fire", "structure fire"], "fire")
        primary_type is the category with the most keyword hits, or None.

    Side-effect: updates calls/{call_id}.tags in Firestore.
    """
    lower = transcript.lower()
    matched: dict[str, list[str]] = {}

    for category, keywords in INCIDENT_KEYWORDS.items():
        hits = [kw for kw in keywords if kw in lower]
        if hits:
            matched[category] = hits

    tags: list[str] = []
    for category, hits in matched.items():
        tags.append(category)
        tags.extend(h for h in hits if h != category)

    # Deduplicate while preserving order
    seen: set[str] = set()
    unique_tags: list[str] = []
    for t in tags:
        if t not in seen:
            seen.add(t)
            unique_tags.append(t)

    # Primary type = category with most keyword hits
    primary_type: Optional[str] = None
    if matched:
        primary_type = max(matched, key=lambda c: len(matched[c]))

    if unique_tags:
        try:
            await fstore.doc_update("calls", call_id, {"tags": unique_tags})
        except Exception as e:
            logger.warning(f"Could not save tags for call {call_id}: {e}")

    logger.info(f"Intelligence: call {call_id} → tags={unique_tags}, type={primary_type}")
    return unique_tags, primary_type


def extract_location_hint(transcript: str) -> Optional[str]:
    """Return the first street-level location mention found in the transcript, or None."""
    match = _STREET_RE.search(transcript)
    return match.group(0) if match else None